[ICLR2026] This is the first paper to explore how to effectively use R1-like RL for MLLMs and introduce Vision-R1, a reasoning MLLM that leverages cold-start initialization and RL training to incentivize reasoning capability.

Official repo of Griffon series including v1(ECCV 2024), v2(ICCV 2025), G, and R, and also the RL tool Vision-R1.

Official implementation of GUI-R1 : A Generalist R1-Style Vision-Language Action Model For GUI Agents

[TMLR 25] SFT or RL? An Early Investigation into Training R1-Like Reasoning Large Vision-Language Models

Med-R1: Reinforcement Learning for Generalizable Medical Reasoning in Vision-Language Models

Code for paper "RealSR-R1: Reinforcement Learning for Real-World Image Super-Resolution with Vision-Language Chain-of-Thought"

[NeurIPS'25] Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding

MobileVLA-R1: Reinforcing Vision-Language-Action for Mobile Robots

We propose IAD-R1, a universal post-training framework that enhances Vision-Language Models for industrial anomaly detection through a two-stage training strategy.

R1-Vision: Let's first take a look at the image

Mitigating Shortcuts in Visual Reasoning with Reinforcement Learning

[ICLR 2026] ReWatch-R1: Boosting Complex Video Reasoning in Large Vision-Language Models through Agentic Data Synthesis

VTC-R1: Vision-Text Compression for Efficient Long-Context Reasoning.

Official implementation of the paper "ETP-R1: Evolving Topological Planning with Reinforcement Fine-tuning for Vision-Language Navigation in Continuous Environments"

Official implementation of "PathReasoner-R1: Instilling Structured Reasoning into Pathology Vision-Language Model via Knowledge-Guided Policy Optimization"

an method to make vlm think like r1

R1V, trained with AI feedback, answers open-ended visual questions.

No description available

One-click start reproduction of multi-modal DeepSeek R1-Zero

This repository include 5 projects for NTI R1 advanced computer vision

official code for the paper "GenSeg-R1: RL-Driven Vision–Language Grounding for Fine-Grained Referring Segmentation"

DocScope-R1 is an experimental, advanced document vision suite designed for high-performance Optical Character Recognition (OCR) and complex visual reasoning.

This is a Multimodal Retrieval-Augmented Generation (RAG) system that extracts text and images from PDFs, retrieves relevant information, and generates responses using DeepSeek-R1-Distill-Qwen-1.5B model for output text generation, BLIP vision model for image captioning and Flet framework for GUI.

This repository contains data published in Lafer-Sousa, R., & Conway, B. R. (2017). #thedress: Categorical perception of an ambiguous color image. Journal of Vision; Lafer-Sousa, R., Hermann, K., L., & Conway, R., B. (2015). Striking Individual Differences in Color Perception Uncovered by 'the Dress' Photograph. Current Biology 25, R1–R2

No description available

基于Tiny-R1架构的极限显存使用场景的GRPO训练框架

[CVPR'26] Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Reward

A privacy-first, fully local Agentic AI legal assistant. Combines multimodal RAG (DeepSeek-R1 + Llama 3.2 Vision) via Telegram for zero-data-leakage contract analysis.

DeepSeek-R1 is a cutting-edge deep learning model designed for advanced image recognition tasks, boasting high accuracy and efficiency in detecting complex patterns within visual data. Leveraging state-of-the-art neural network architecture, DeepSeek-R1 sets a new benchmark in the field of computer vision by pushing the boundaries of what is possib

Multi-modal AI video editing pipeline. Whisper ASR + Qwen2-VL vision + DeepSeek-R1 reasoning → FCPXML timelines for DaVinci Resolve. Features 33:1 metadata compression, semantic vector search, and A-roll/B-roll classification. IIT Ropar Module E project.